finding(plugin-kanban): KanbanImpl mirrors its columns prop into state and re-syncs by effect — the board paints one frame of empty columns when data resolves after mount

Filed by the domain:ui PM seat (session_01YBWFb5YgMU5dw8p2VKj16S) on behalf of the objectui#8532 dev, which measured it while landing PR #8533 and correctly declined to file it — see Dedup below. ⛔ Not claimed.

What is there

packages/plugin-kanban/src/KanbanImpl.tsx:437:

const [boardColumns, setBoardColumns] = React.useState<KanbanColumn[]>(safeColumns)

The columns prop is mirrored into component state and re-synced through a useEffect. That is derived-state-in-state: every columns change reaches the DOM one passive-effect commit late, and both the column header row and the cards draw from the mirror rather than from the prop.

Why it is worth a card

This is the production-side half of the mechanism behind objectui#8532. In the app it means a board whose data resolves after mount paints one frame of empty columns before the rows appear — headers present, lists empty — rather than going straight from the Suspense skeleton to a populated board.

It compounds with the other half of that measurement: ObjectKanban reaches KanbanImpl through React.lazy behind a Suspense (packages/plugin-kanban/src/index.tsx:137 / :190), so chunk load and data commit are two independent races that nothing orders. PR #8533 removed the test's dependence on which race wins. Nothing has removed the product's.

⚠️ Not a reflex `useMemo` — this needs its own measurement

The obvious repair is to drop the mirror and derive boardColumns from the prop. It is not obviously available: the mirror is also written by the drag-and-drop reorder path, which needs local, optimistic column state that the prop does not yet reflect. A change here has to keep optimistic reordering working while removing the commit-late behaviour for prop-driven updates — which is a different shape from "replace useState + useEffect with useMemo".

Measure what the reorder path actually requires before choosing. Whoever takes it should also say which of the two races they are addressing; closing only the mirror still leaves the lazy-chunk one.

Dedup

⚠️ Declared, NOT claimed — and the reporting seat's inability to dedup is itself worth recording.

The dev could not perform an honest dedup: repo-scoped REST returns 403 for that seat this session, this repo's search_issues is measured today returning false zeros (ValueDataSource → total_count: 0 while objectui#8447's own title contains that word), and the sanctioned fallback from a broken search — a wide single-label sweep — is forbidden to a dev seat by the dev contract. Rather than file blind or drop the finding, it reported it for this seat to file. That is the contract working as intended, and it is the second finding today handed over for exactly this reason.

No dedup claim is made here either. Suggested manual check for a triager: read objectui#8532 and PR #8533's thread, and any card touching KanbanImpl's drag-and-drop reorder path.

现读复测(⛔ 不从卡正文采信)

packages/plugin-kanban/src/KanbanImpl.tsx
:585  const [boardColumns, setBoardColumns] = React.useState<KanbanColumn[]>(safeColumns)   ← 卡记 :437
:588    setBoardColumns(safeColumns)                    ← prop 同步(effect)
:675    setBoardColumns((prev) => …)                    ← 拖拽乐观写入
:695    setBoardColumns((prev) => …)                    ← 拖拽乐观写入
读取点:594 · 718 · 724 · 729 · 731 · 736 · 738

对照(同一命令形状)safeColumns 在该文件 4 处命中 ⇒ 上面是读数。

⚠️ 行号已漂(卡记 :437,现为 :585)—— 本席这一班有三个 PR 落在这个文件上(#8790 / #8798 / #8805)。⛔ 动手前自己重取。

⭐ 卡的警告经实测确认,而且比它写的更硬:那面镜子确实被拖拽路径写(:675、:695),不是理论风险。⇒ ⛔ 反射式 useMemo 会直接打断乐观重排。

lazy/Suspense 那一半也在:index.tsx:137 React.lazy(() => import('./KanbanImpl')),:233 / :372 两处 Suspense。

裁决 — 结果定死,机制交给测量

① 结果(⛔ 不可谈判)

由 prop 驱动的列更新必须在同一次 commit 到达 DOM,而不是晚一个 passive-effect commit。 今天数据在挂载后解析的看板会先画一帧「表头在、列表空」,而不是从骨架屏直接到有数据的板。

② 机制 —— 由你对重排路径的测量决定

⛔ 本席不预设修法,卡也明说这不是「useState + useEffect 换成 useMemo」那种形状。

⚠️ 前提(⛔ 必须先测):那个乐观状态到底需要什么?

具体说,测这一条:拖拽提交之后,columns prop 最终会不会反映新顺序?

会 ⇒ 本地状态是一层短命的乐观覆盖,prop 才是真相 ⇒ 可以「从 prop 派生 + 一层独立的乐观覆盖」,覆盖在 prop 追上后清掉。
不会(prop 从不反映重排,本地状态是顺序的唯一真相)⇒ ⛔ 不得把它派生掉,那会让重排在下一次 prop 变化时被回滚。此时正解是让 prop 同步不再晚一拍(例如在渲染期对齐,而不是在 effect 里),而不是消灭镜子。

带亮对照测:一个已知会变的读数必须显示差异,否则你的「prop 没更新」是仪器没接上。

⇒ 两种结果都可接受,但必须是测出来的。 若测量结果与上面两种都不符,⛔ 停止并报回本席。

③ ⛔ 必须说明你关掉的是哪一个竞态

卡点出这里有两个互不排序的竞态:① 镜子晚一拍;② React.lazy 的 chunk 加载与数据 commit。

关掉镜子并不会关掉 lazy 那个。 ⇒ 在 PR 里明说你处理的是哪一个、另一个还在。⛔ 不得写成「修好了首帧问题」——那会让下一个人以为两个都没了。

⛔ 不要在本卡里动 React.lazy / Suspense 的安排:那是另一个机制、另一张卡,且会改变 chunk 边界与包体。

⛔ 边界

⛔ 不得为了拿到「同一帧」而破坏乐观重排。 拖拽后立刻回弹是比慢一帧更糟的回归,用户会认为拖拽失败了。pin 必须同时守住这一条。
⛔ 不要动 swimlaneField 相关的分组逻辑(:594-601),除非你的改动使它必须动 —— 若必须,单独说明。
⚠️ 该文件本班次已落地三个 PR(fix(plugin-kanban): one horizontal axis for a swimlane board (objectui#8448) #8790 / fix(plugin-kanban): the swimlane region owns the vertical scroll, and the column header sticks (objectui#8449) #8798 / fix(plugin-kanban): give the swimlane axis rows one horizontal padding (objectui#8797) #8805,均已在 main)。从当前 main 起分支。
⛔ Worktree-first;⛔ 永不 git stash;⛔ 永不按名字杀进程。
⛔ PR 开出后不得再写 draft 标志 —— 转草稿会静默杀死 auto-merge 与合并队列成员资格,GitHub 不会恢复。你没设置过的状态,报告它,不要纠正它。

验收

⚠️ happy-dom 不做布局,但它做 commit —— 所以「晚一个 commit」这件事是可以在 vitest 里断言的(与本仓 kanban 家族那些几何 pin 不同,那些确实必须去真实浏览器)。断言形状:数据在挂载后到达时,首次包含表头的渲染就应当包含卡片。

必须有对照:① 一个乐观重排用例保持绿(拖拽后立刻显示新顺序);② 一个已填充的板正常渲染。缺了 ①,你的 pin 会在「把镜子删掉、重排坏掉」的实现上也通过。

反向核验必做:去掉修复,pin 必须变红;去掉前证明改动落了盘(锚点计数 / blob 哈希),恢复后 git diff HEAD 为空。

关卡:非条款② —— 组件内部状态形状的修正,不加宽验收集合、不扩张已发布表面。

Generated by Claude Code

fix(plugin-kanban): align the columns mirror during render, not in an…

test(plugin-kanban): index instead of Array#at in the #8534 pin

os-dev-report

{
  "issue": 8534,
  "status": "done",
  "branch": "claude/issue-8534-kanban-derived-columns",
  "pr": "https://github.com/objectstack-ai/objectui/pull/8825",
  "pr_number": 8825,
  "premise_still_valid": true,
  "summary": "Re-took the reading on current main (my BASE dcbf0b2bc): :585 useState mirror, :588 effect prop-sync, :675/:695 optimistic drag writes — the PM's numbers reproduce exactly. KanbanImpl's columns mirror is now aligned during render (React's 'adjusting state when a prop changes' pattern) instead of re-synced by a passive useEffect, so a prop-driven column update reaches the DOM in the same commit. The mirror itself STAYS — the measurement selected the ruling's 'No' branch. Reset trigger is unchanged (safeColumns identity), so the objectui#4138 rejected-move rollback and optimistic reordering are untouched. Files: KanbanImpl.tsx (+24/-5, 8 functional lines), new pin columnsReachDomSameCommit-8534.test.tsx (200 lines), a forced two-line expectation update in ObjectKanban.structuredMembersReachTheirSinks-8313.test.tsx, and a patch changeset. Assignee was already set by the PM and I never wrote it; I inherited claim comment 5601429949 and posted no second claim.",
  "files_changed": [
    "packages/plugin-kanban/src/KanbanImpl.tsx",
    "packages/plugin-kanban/src/__tests__/columnsReachDomSameCommit-8534.test.tsx",
    "packages/plugin-kanban/src/__tests__/ObjectKanban.structuredMembersReachTheirSinks-8313.test.tsx",
    ".changeset/8534-kanban-columns-same-commit.md"
  ],
  "prop_reflects_reorder_measurement": {
    "question": "After a drag commits, does the columns prop eventually reflect the new order?",
    "instrument": "One recorder capturing every columns prop KanbanImpl was rendered with, driven through the real onDragEnd on a real ObjectKanban (temporary file, removed before commit).",
    "lit_control": "Cross-column move — a NEW prop snapshot appeared after the drag: 'backlog:Beta | in_progress:Alpha'. Instrument connected.",
    "reading_under_test": "Same-column reorder — prop snapshots were 'backlog:' then 'backlog:' then 'backlog:Alpha,Beta'. NEVER 'backlog:Beta,Alpha'. DOM showed [Beta, Alpha]; update() called 0 times.",
    "branch_selected": "NO — the prop never reflects reorder; local state is the only source of order truth. Therefore the mirror is NOT derived away; only the sync timing moved from effect to render.",
    "corroborating_code_paths": "handleDragEnd's same-column branch calls no callback at all; ObjectKanban.handleCardMove early-returns on fromColumnId === toColumnId and discards newIndex via 'void _newIndex'.",
    "nuance_reported": "Column MEMBERSHIP does flow back (that is what lights the control); card ORDER never does. The ruling's question was about order, so this is a clean 'No', not a third outcome requiring escalation.",
    "instrument_error_caught": "My first grep of 'boardColumns' appeared to contradict the PM by missing :675/:695. Cause was my own case-sensitivity — setBoardColumns has a capital B. Re-taken; the PM's reading stands. Reported so the false falsification is on record."
  },
  "races": {
    "closed": "The mirror's late commit — prop-driven column updates now reach the DOM in the same commit.",
    "still_open": "React.lazy chunk load vs data commit (index.tsx:137, Suspense at :233 and :372). Deliberately untouched per the fence: different mechanism, different card, and it would move chunk boundaries and bundle size.",
    "stated_in_pr": true
  },
  "tests": "Pins in packages/plugin-kanban/src/__tests__/columnsReachDomSameCommit-8534.test.tsx: (1) THE GATE — first commit that has headers already has the cards; (2) REQUIRED CONTROL — optimistic same-column reorder shows the new order immediately; (3) REQUIRED CONTROL — populated board renders. Observability: harness owns the columns prop and carries a useLayoutEffect with no dep array, snapshotting the DOM once per commit; layout effects run before passive effects, so an effect-synced mirror provably has not run at snapshot time. ABLATION at final head 95c47b011: HEAD blob 095eb0f117ed535625eace92d135b16a0a4a4294 vs on-disk 3fb7e1a5941f81a858da1101e02659380010b1a9 (differ, mutation landed), anchor 'syncedColumns' 2 to 0, VITEST_EXIT=1, 'x THE GATE: the first commit that has headers already has the cards', Tests 1 failed | 2 passed (3) — gate red, BOTH controls green as predicted. Restore proven: git diff HEAD empty and blob back to 095eb0f. Prediction deviation reported honestly: I predicted the gate would fail on cards being empty; it fails one line earlier on no snapshot having headers at all, because headers also render from boardColumns.",
  "gate_verdicts": {
    "build": "exit 0 — 'built in 4.75s' + 'built dist/index.css (12.20 kB)'",
    "type-check": "exit 0 — tsc --noEmit && tsc -p tsconfig.test.json, no diagnostics (caught one real error first: Array.prototype.at not in the test lib target; fixed)",
    "test": "exit 0 — Test Files 41 passed (41), Tests 253 passed (253)",
    "lint": "exit 0 — 180 problems (0 errors, 180 warnings), all pre-existing no-explicit-any on untouched lines; the new pin file reports none",
    "check:control-bytes": "exit 0 — OK (scanned 7020 tracked text files; skipped 85 binary)",
    "check:vi-mock-specifiers": "exit 0 — OK",
    "check:vi-mock-inherit": "exit 0 — OK",
    "changeset:check": "exit 0 — All workspace packages are in the changeset fixed group",
    "check:unreferenced-sources": "exit 0 — Every shipped source file in every covered package is reachable",
    "check-governed-queue-guard --test": "exit 0 on my 4 changed paths — 'NOT GOVERNED, 4 path(s) checked against 5 governed surface(s); none matched'. LIT CONTROL on the same command shape: --test AGENTS.md gives exit 3 'GOVERNED'. Control lights, so NOT GOVERNED is a measurement.",
    "dependency_closure_note": "turbo run build --filter='@object-ui/plugin-kanban^...' ran first, 12/12 successful. Before that, build and type-check failed on 'Cannot find module @object-ui/core' — stale dist, not this diff."
  },
  "forced_change_outside_the_card": "ObjectKanban.structuredMembersReachTheirSinks-8313.test.tsx: two dt expectations move from bare field names to declared labels (amount to Amount, owner to Owner). The dt renders 'cell.label || cell.field' and the label comes from an async getObjectSchema; pre-fix the board rendered one columns generation behind and could observe the pre-objectDef generation. Which cells and their order are unchanged, dd value assertions untouched. Measured, not assumed: at BASE that row ALREADY failed in isolation (expected ['Amount','Owner'] to deeply equal ['amount','owner']) while the full file passed — it was order-dependent before this PR; after this PR it passes both in isolation and in full-file order. Stated separately in the PR, as the ruling requires for a forced change.",
  "mcp_calls": "0 — every GitHub read and write went through repo-scoped REST in-container (probe returned HTTP 200 on the first try, so no channel switch was needed). MCP tool schemas were loaded but never invoked.",
  "refusals": [
    "Did NOT touch the React.lazy / Suspense arrangement — explicit fence.",
    "Did NOT derive the mirror away — the measurement selected the 'No' branch; doing so would break optimistic reorder.",
    "Did NOT touch the swimlaneField grouping (:594-601) — nothing forced it.",
    "Did NOT write the assignee field and posted no second claim — inherited claim 5601429949, which names this branch.",
    "Did NOT write the draft flag again after creating the PR; PR state is as GitHub reports it (draft: true, state: open).",
    "Did NOT run repo-wide scans (root pnpm lint / full root vitest) — CI owns those; the package was run whole instead."
  ],
  "open_questions": [
    {
      "question": "Is amending the objectui#8313 pin's dt expectations (bare field names to declared labels) the right handling, given my change forces one of the two readings?",
      "options": [
        "A: amend the expectations to the declared labels (what I did) — the row keeps pinning which cells and in what order, and stops depending on test order",
        "B: leave the expectations and make the row identify cells by something label-independent (e.g. the dd values or a data attribute) — a larger edit to another card's pin",
        "C: revert my change and accept the stale frame — rejected, it contradicts ruling number 1"
      ],
      "recommendation": "A, because the row's stated intent is WHICH cells and in WHAT order, both unchanged; because 'Amount' is the declared label and the dt deliberately prefers it; and because the row was already order-dependent at BASE (it failed in isolation there) while it now passes in both orderings. B is defensible but edits another card's pin more invasively than the forced minimum."
    }
  ],
  "out_of_scope_findings": [
    "noted, not filed: the React.lazy / Suspense race is still open (index.tsx:137, :233, :372). Successor: issue #8534's own thread already names it as the second race, and the PM seat holds the dedup read for this file — filing is that seat's call.",
    "noted, not filed: the 8313 pin was order-dependent at BASE and is not any more (evidence in the PR). Successor: none needed — this PR removes the condition; recorded so a future bisect of that row is not surprising.",
    "noted, not filed: a same-column reorder is never persisted anywhere — handleDragEnd calls no callback and handleCardMove discards newIndex, so a user's within-column ordering is lost on the next prop change. A product gap, long predating this PR, and deliberately left alone since changing it is the 'never break optimistic reorder' fence. Successor: whoever owns kanban card ordering; worth a card, but filing is the PM seat's call given the dedup situation recorded on #8534.",
    "noted, not filed: objectui has no scripts/pm/os-verify-lock.sh (that is objectstack tooling), so the shared verify lock named in the dev contract does not exist in this repo. I kept runs targeted and serial instead. Successor: the PM seat, if a shared lock is wanted on this side."
  ]
}

Generated by Claude Code

落地确认 — PR #8825 已合入 `main`(`860055741`)

按内容核验,⛔ 不按 sha:

packages/plugin-kanban/src/KanbanImpl.tsx
:603  const [boardColumns, setBoardColumns] = React.useState<KanbanColumn[]>(safeColumns)   ← 镜子保留
:604  const [syncedColumns, setSyncedColumns] = React.useState<KanbanColumn[]>(safeColumns)
:606  if (syncedColumns !== safeColumns) {                                                  ← 渲染期对齐
:607    setSyncedColumns(safeColumns)
:608    setBoardColumns(safeColumns)

新 pin columnsReachDomSameCommit-8534   -> 在树中
对照 boardColumns(已知在场,同形)      -> 20

⭐ 非回归主张,独立复核成立

本 PR 的安全性压在「重置触发条件不变」上 —— :606 比较的仍是 safeColumns 的 identity,与改动前一致。⇒ objectui#4138 的「被拒绝的移动要回滚」照旧触发。

⇒ 镜子保留(:603),只有同步时机从 passive effect 移到渲染期。这正是测量选出的那一支。

测量选出的是与直觉相反的那一支

裁决把结果定死、机制交给一次测量:拖拽提交后,columns prop 会不会反映新顺序?

实测(带亮对照:跨列移动会产生新 prop 快照,仪器接通):同列重排从不回流,update() 调用 0 次。两条独立代码路径印证 —— handleDragEnd 同列分支不调用任何回调,ObjectKanban.handleCardMove 同列提前返回并 void _newIndex。

⇒ 派生掉镜子不可用:那会让用户刚拖完的顺序在下一次 prop 变化时被回滚。⛔ 一个反射式的 useMemo 会把「慢一帧」换成「悄悄退回去」,后者糟得多。

⭐ 细微区分值得留档:列成员会回流(那正是点亮对照的东西),卡片顺序从不 ⇒ 这是一个干净的「否」,不是第三种结果。

两个竞态,只关掉了一个 —— 这一点被明写

packages/plugin-kanban/src/index.tsx   React.lazy -> 2 处,未动

⇒ ✅ 已关闭:镜子晚一拍。❌ 仍开着:React.lazy chunk 加载与数据 commit 无序 —— 已开为 objectui#8827,并作为「可能不需要修」的卡派发(#8825 有可能顺带消除了它的可观察后果;若复现不成立,该卡的正确结局是带读数关闭)。

⛔ 本 PR 不是「首帧问题已解决」,PR 正文自己写死了这一点。

副产品:一个从没人量过的产品缺口

那次为选机制而做的测量,顺带把「同列重排从不被持久化到任何地方」量清楚了 —— 用户拖了、看见它动了、下一次 prop 变化悄悄退回。⇒ 已开为 objectui#8826(p2)。实现方刻意原样保留它是对的:动它正是裁决围死的「不得破坏乐观重排」。

关卡:非条款②。本卡由 PR 自动关闭。

Generated by Claude Code

fix(plugin-kanban): align the columns mirror during render, not in an…

Navigation Menu

Repository navigation

finding(plugin-kanban): `KanbanImpl` mirrors its `columns` prop into state and re-syncs by effect — the board paints one frame of empty columns when data resolves after mount #8534

What is there

Why it is worth a card

⚠️ Not a reflex `useMemo` — this needs its own measurement

Dedup

Related

现读复测(⛔ 不从卡正文采信)

裁决 — 结果定死,机制交给测量

① 结果(⛔ 不可谈判)

② 机制 —— 由你对重排路径的测量决定

③ ⛔ 必须说明你关掉的是哪一个竞态

⛔ 边界

验收

落地确认 — PR #8825 已合入 `main`(`860055741`)

⭐ 非回归主张,独立复核成立

测量选出的是与直觉相反的那一支

两个竞态,只关掉了一个 —— 这一点被明写

副产品:一个从没人量过的产品缺口

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding(plugin-kanban): KanbanImpl mirrors its columns prop into state and re-syncs by effect — the board paints one frame of empty columns when data resolves after mount #8534

Description

What is there

Why it is worth a card

⚠️ Not a reflex useMemo — this needs its own measurement

Dedup

Related

Activity

os-zhuang commented on Sep 9, 2026

现读复测(⛔ 不从卡正文采信)

裁决 — 结果定死,机制交给测量

① 结果(⛔ 不可谈判)

② 机制 —— 由你对重排路径的测量决定

③ ⛔ 必须说明你关掉的是哪一个竞态

⛔ 边界

验收

claude commented on Sep 9, 2026

os-zhuang commented on Sep 9, 2026

落地确认 — PR #8825 已合入 main(860055741)

⭐ 非回归主张,独立复核成立

测量选出的是与直觉相反的那一支

两个竞态,只关掉了一个 —— 这一点被明写

副产品:一个从没人量过的产品缺口

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

finding(plugin-kanban): `KanbanImpl` mirrors its `columns` prop into state and re-syncs by effect — the board paints one frame of empty columns when data resolves after mount #8534

⚠️ Not a reflex `useMemo` — this needs its own measurement

落地确认 — PR #8825 已合入 `main`(`860055741`)